How to optimize your AI token usage

A new open-source tool called Repo-Brain compresses entire codebases into a single markdown context file, achieving 96% compression on a 262-file repository by reducing 154,229 tokens to 6,487. The tool uses static analysis, architecture analysis, and semantic relationship mapping to feed code context to any large language model in a single prompt, eliminating the need to re-read repositories during each conversation.

Compress an entire codebase into a single markdown context file. Feed it to any LLM once instead of re-reading your repo every conversation. Achieved 96% compression on a 262-file repo 154,229 → 6,487 tokens . What's included Static analysis — Tree-sitter AST parsing for Python, JS, TS, Go, Rust; regex fallback for Java, Ruby, C , C/C++, Swift, Kotlin, Shell, and more Architecture analysis — single LLM call identifies layers, components, entry points, and data flow Semantic relationships — LLM-discovered producer/consumer links, shared data structures, parallel implementations, and polyglot bridges Multi-provider support — OpenAI, Claude, Deepseek, Gemini, Groq, Ollama, Mistral, xAI, Perplexity, OpenRouter One-liner installers — no manual venv or config setup required Install Mac / Linux: curl -fsSL https://github.com/KrishivPiduri/repo-brain/releases/latest/download/install.sh https://github.com/KrishivPiduri/repo-brain/releases/latest/download/install.sh | bash Windows PowerShell : irm https://github.com/KrishivPiduri/repo-brain/releases/latest/download/install.ps1 https://github.com/KrishivPiduri/repo-brain/releases/latest/download/install.ps1 | iex Assets Upload these files to this release: - install.sh - install.ps1 - repo-brain.zip zip of: main.py, llm.py, ingest.py, analyze.py, relationships.py, generate prompt.py, mcp server.py, config.example.py, requirements.txt